module Data.ANN.Hopfield where

import Data.List

-- Discrete Hopfield networks

-- Hopfield networks are a form of error-correcting, auto-associative memory.
-- That is, a hopfield network memorises a list of vectors and can often find the closest
-- memorised vector for a given input.

-- Quick and dirty version using lists. Would be nice to do this using a vector space library

type Vector a = [a]
type Matrix a = [Vector a]

vecPlus = zipWith (+)
matPlus = zipWith vecPlus

matId i = (i : repeat 0) : (map (0:) $ matId i)
matProd xs ys = [ [sum $ zipWith (*) x y | y<-transpose ys] | x<-xs]

outerProd xs ys = [ [x*y | y<-ys] | x<-xs]

boolToInt b = if b then 1 else -1
intToBool i
    | i >= 0    = True
    | otherwise = False

-- A hopfield network can be represented as a matrix multiplication
data Hopfield = Hopfield (Matrix Int)
    deriving (Eq,Ord,Show)

makeHopfield :: [Vector Bool] -> Hopfield
makeHopfield bs = Hopfield $ foldl matPlus (matId (- genericLength bs)) [ y `outerProd` y | y <- map (map boolToInt) bs]

recall :: Hopfield -> Vector Bool -> Vector Bool
recall (Hopfield ys) x = [intToBool $ sum $ zipWith (*) (map boolToInt x) y | y<-ys]